Preamble and Founding Principles
The basis and intent of this charter
The EM Foundation is a Texas nonprofit corporation organized for the purpose of advancing the responsible development and deployment of artificial intelligence through independent assessment, public research, and governance standard-setting. The Foundation operates no AI systems of its own. It has no commercial interest in the success or failure of any AI system it assesses. It exists to serve the public interest — specifically the interest of people who interact with AI systems and who deserve accurate, independent information about the systems they use.
The Foundation recognizes that independent assessment of AI systems is a public good that no commercial actor can adequately provide for itself, and that no government body can provide with sufficient technical agility. It also recognizes that organizations claiming independence can be captured — by the industries they assess, by the funders who sustain them, by the political pressures they face, and by their own institutional interests. This charter is designed to make capture visible, difficult, and correctable.
The Foundation draws explicitly on the traditions of Consumer Reports (independent product testing funded without advertiser relationships), NIST (technical standard-setting in the public interest), ISO and IEC (international consensus standards with defined revision processes), the institutional design of the Cochrane Collaboration (systematic review with transparency and conflict requirements), and the governance structure of credit rating reform proposals following the 2008 financial crisis (independence through structural separation from rated entities).
Nothing in this charter makes the Foundation infallible. Scores will contain errors. Methodologies will have limitations. Reviewers will occasionally fail. This charter is designed not to prevent all failure but to make failures visible, correctable, and attributable — and to make capture structurally difficult rather than merely prohibited by policy that is impossible to enforce.
Founding Principles
The following principles are interpretive guides for all provisions of this charter. When a provision is ambiguous, the interpretation that best advances these principles governs.
PRIN-001Independence is structural, not declaratory. The Foundation's independence is not established by a statement that it is independent. It is established by structures that make dependence on any single actor economically and procedurally impossible. Declarations of independence that are not backed by structural mechanisms are not independence.
PRIN-002Transparency is the primary accountability mechanism. Because the Foundation operates without market competition and with limited external enforcement, public transparency — of methodology, funding, conflicts, and decisions — is the primary mechanism by which the public can hold the Foundation accountable. Transparency is not a governance option; it is the governance system.
PRIN-003The assessor must have nothing to gain from the assessment outcome. Every structural provision governing assessors and reviewers derives from this principle. Any relationship that gives an assessor a financial, professional, or reputational stake in a specific assessment outcome is a conflict of interest regardless of the assessor's subjective intent.
PRIN-004Procedural fairness is owed to assessed entities. AI developers whose systems receive published scores have legitimate interests in accuracy, fair process, and opportunity to respond. Procedural fairness to assessed entities is not in tension with independence — it is a component of the methodological integrity that makes assessments credible.
PRIN-005The public interest is the ultimate principal. When the interests of the Foundation's funders, the preferences of assessed providers, the convenience of the Board, and the public interest conflict, the public interest governs. The Foundation exists to serve people who use AI systems. All governance structures ultimately answer to them.
PRIN-006Capture is the failure mode, not incompetence. The Foundation anticipates that competent, well-intentioned actors will make errors. It does not primarily anticipate incompetence; it primarily anticipates capture — the gradual alignment of the Foundation's outputs with the interests of powerful actors rather than with the public interest. Every major governance provision in this charter is designed to make capture visible, difficult, or recoverable.
PRIN-007Institutional courage is an organizational obligation. The Foundation will publish findings that harm powerful actors' commercial interests. It will maintain positions under litigation pressure where those positions are methodologically sound. It will correct errors publicly even when doing so is embarrassing. Institutional courage is not a personality trait of individual staff — it is a structural commitment backed by financial reserves, legal preparation, and governance protections that make it possible.
Article I — Name, Purpose, and Scope
Identity and jurisdictional limits of this charter
§ 1.1Legal name. The full legal name of the organization governed by this charter is EM Foundation, a Texas nonprofit corporation. The organization may operate under the abbreviated name "EM Foundation" for all public-facing purposes.
§ 1.2Organizational purpose. The Foundation is organized and operated exclusively for charitable and educational purposes within the meaning of Section 501(c)(3) of the Internal Revenue Code. Its specific purposes are: (a) conducting independent assessment of artificial intelligence systems against published governance frameworks; (b) developing, publishing, and maintaining technical and governance standards for AI assessment; (c) educating the public, policymakers, and industry on AI system behavior, capabilities, and limitations; and (d) supporting academic and civil society research on AI governance.
§ 1.3Assessment authority. This charter governs all assessment activities conducted under the Intelligence Assessment Framework (IAF) and all subsequent versions thereof, all activities governed by the Corroboration Standard and its successors, all activities associated with the AI Assessment Index and any successor public-facing assessment publication, and any additional assessment programs formally adopted by Board resolution.
§ 1.4What the Foundation is not. The Foundation is not a regulatory body and holds no governmental authority. Published scores are assessments of AI system behavior against disclosed criteria at the time of assessment — they are not safety certifications, legal compliance determinations, or endorsements. The Foundation is not a consumer protection agency and does not represent individual users in disputes with AI providers. The Foundation is not a lobbying organization and does not advocate for specific legislation or regulatory outcomes, except in its capacity as a source of technical information for policymakers.
§ 1.5International aspiration. The Foundation intends its assessment frameworks and standards to be applicable to AI systems deployed globally. It will seek alignment with and recognition from international standards bodies including ISO, IEC, and national standards bodies. The Foundation's governance structures are designed to meet or exceed the independence and transparency requirements of recognized international standards organizations.
Article II — Governance Structure
Organizational bodies and their relationships
The Foundation's assessment governance operates through five distinct bodies with separated authorities. No single body holds authority over the full assessment lifecycle. This separation is a structural anti-capture mechanism, not an administrative convenience.
2.1 Board of Directors
The Board of Directors holds ultimate fiduciary authority over the Foundation. It sets organizational policy, adopts and amends this charter, appoints the Executive Director, approves major funding decisions, and serves as the final internal body for appeals. The Board does not conduct assessments, score systems, or make individual assessment decisions. Board members who hold roles in the assessment process must recuse from Board decisions related to those specific assessments.
2.2 Assessment Committee
The Assessment Committee is a standing operational committee responsible for oversight of all active assessments. It approves assessment protocols for specific assessment cycles, monitors compliance with this charter and the published methodology, and approves the publication of scores. The Committee consists of the Director of Assessment (a senior staff role), two Board members who meet Assessment Independence requirements (§ 4.2), and two external subject matter experts appointed by the Board for renewable two-year terms. The Assessment Committee does not conduct assessments directly — it oversees the assessors who do.
2.3 Independent Review Panel
The Independent Review Panel (IRP) is a standing body of domain experts responsible for adjudicating disputes that reach Stage 2 or Stage 3 under Article XI. IRP members are not employed by the Foundation; they are contracted for specific dispute resolution engagements. IRP members must meet Assessment Independence requirements and must not have served as assessors for the specific assessment under dispute. The IRP has no authority over assessments that are not in active dispute. IRP composition, appointment, and rotation are governed by Schedule A of this charter.
2.4 Scientific Advisory Panel
The Scientific Advisory Panel (SAP) is a standing body of academic and technical experts that advises on methodological questions, makes binding determinations on contested settled/non-settled scientific question classifications (§ 6.8), and contributes to methodology revision processes under Article XIV. SAP members serve in an advisory and adjudicative capacity only — they do not conduct assessments and do not hold Board authority. SAP composition and appointment are governed by Schedule B.
2.5 Permanent Adversarial Function
The Permanent Adversarial Function (PAF) is a standing function — staffed internally or through contracted arrangement — dedicated to actively attempting to identify weaknesses in the Foundation's governance mechanisms, methodology, and operational security. The PAF reports quarterly to the Board, independently of Foundation management. Its reports are published publicly within 30 days of Board receipt, with redaction only for information whose publication would itself enable attacks. The PAF is the institutional operationalization of the principle that a serious governance organization must actively search for its own vulnerabilities. The PAF charter, scope, and reporting requirements are governed by Schedule C.
2.6 Authority hierarchy
| Body | Authority | Cannot Do |
|---|
| Board of Directors | Organizational policy; charter; budget; personnel; final appeals; charter amendment | Conduct assessments; override specific assessment scores (except through defined correction process); earmark funds for specific assessments |
| Assessment Committee | Assessment protocol approval; score publication approval; methodology compliance oversight | Modify individual scores; override reviewer determinations without defined process; approve funding decisions |
| Independent Review Panel | Binding dispute resolution; Stage 3 appeal decisions | Initiate assessments; modify methodology; approve scores outside dispute context |
| Scientific Advisory Panel | Settled/non-settled question determinations; methodology consultation | Approve or deny publication of specific scores; conduct assessments |
| Permanent Adversarial Function | Governance vulnerability identification; quarterly public reporting | Modify governance structures; approve or block assessments |
Article III — Board of Directors
Composition, qualifications, terms, and conduct
3.1 Size and composition
§ 3.1.1The Board shall consist of not fewer than seven (7) and not more than eleven (11) directors, including the Chair. The Board shall set and maintain its own size by resolution within this range. A Board that falls below seven directors due to vacancy must fill vacancies within ninety (90) days or the remaining directors must suspend assessment score publication until composition requirements are restored.
§ 3.1.2Independence supermajority. At all times, at least seven-ninths (7/9) of the Board — or, if the Board has fewer than nine members, at least two-thirds — must be Assessment Independent as defined in § 4.2. If this supermajority is lost due to vacancy or loss of Assessment Independence status, the Board may continue to operate with reduced quorum requirements for routine matters but may not approve the publication of new assessment scores until the supermajority is restored.
§ 3.1.3Sector concentration limit. No single sector — defined as AI development and deployment; government and regulatory; academic and research; civil society and public interest; legal and professional services; journalism and media — may hold more than two (2) seats on a nine-member Board or a proportionate fraction for boards of other sizes. This limit prevents any single professional community from dominating Board culture regardless of individual independence.
§ 3.1.4Required representation. At all times the Board must include at least: one (1) director with substantial expertise in AI or machine learning systems; one (1) director from academic or research background; one (1) director representing civil society, consumer advocacy, or public interest; and one (1) director with legal or regulatory expertise. These seats are not designated — they are minimum competency requirements for collective Board composition.
3.2 Terms, rotation, and succession
§ 3.2.1Directors serve terms of three (3) years. Terms are staggered such that no more than one-third of the Board's seats expire in any single year. No director may serve more than two (2) consecutive full terms. After a gap of at least three (3) years, a former director may be re-elected for one additional consecutive term.
§ 3.2.2The Chair is elected by Board majority from among the Assessment Independent directors and serves a two-year renewable term as Chair, independent of their term as director. The Chair may not simultaneously hold any compensated role at an AI developer or AI-adjacent commercial entity.
3.3 Director qualifications and disqualifications
§ 3.3.1Directors must: have demonstrated expertise in a field relevant to the Foundation's mission; be capable of exercising independent judgment; commit to the time requirements of Board service (estimated minimum twelve hours per month); and complete the Foundation's orientation on assessment methodology, governance obligations, and conflict-of-interest requirements within sixty (60) days of election.
§ 3.3.2Automatic disqualification events. A director is automatically disqualified and their seat immediately vacated if they: accept employment at an AI developer while serving; acquire equity in an AI system provider above a de minimis threshold of $5,000 in aggregate; fail to disclose a material conflict of interest and the non-disclosure is subsequently discovered; are convicted of a felony or financial crime; or are found to have knowingly participated in any attempt to improperly influence an assessment outcome.
3.4 Meetings and quorum
§ 3.4.1The Board shall meet at minimum four (4) times per year. A quorum consists of a majority of the seated directors. Major decisions — defined as: charter amendment, approval of annual budget, termination of the Executive Director, adoption of new assessment frameworks, and suspension of published scores pending correction — require a two-thirds supermajority of seated directors.
§ 3.4.2Board meeting minutes, including votes and the basis for major decisions, are published on the Foundation's website within thirty (30) days of meeting, with redaction only for personnel matters and legally privileged communications. The practice of holding governance decisions in private and publishing only conclusions is inconsistent with this charter's transparency requirements.
3.5 Compensation
§ 3.5.1Directors may receive reasonable compensation for Board service, approved annually by the Board and published in the Foundation's annual transparency report. Director compensation must be reasonable relative to comparable nonprofit governance roles and must not create financial dependency on Foundation service. No director may receive compensation from any assessed AI provider during their term.
Article IV — Independence Safeguards
Structural separation from assessed entities
Assessment Independence is the foundational independence standard in this charter. A person is Assessment Independent if they have no financial relationship with any AI developer, AI system provider, or entity that has been or may foreseeably be assessed under Foundation frameworks, within the preceding thirty-six (36) months. Financial relationships that disqualify include: employment or consulting arrangements; equity ownership above $5,000; board service (paid or unpaid); advisory roles with compensation; and grants from a person or entity to an organization the person controls or significantly influences. Assessment Independence is assessed at enrollment and re-certified annually.
4.1 Assessor independence
§ 4.1.1All assessors — persons who conduct assessments, score AI system outputs, or contribute to the determination of any published score — must be Assessment Independent with respect to the specific system being assessed. The Foundation maintains a registry of all active assessors and their disclosed financial relationships, updated at minimum annually and whenever a material change occurs.
§ 4.1.2Rotating assessor pools. No assessor may assess the same AI provider's systems in two consecutive assessment cycles. Assessment cycles are defined as annual periods. This rotation requirement applies regardless of the assessor's subjective confidence in their independence and regardless of the provider's preference for assessors familiar with their systems.
§ 4.1.3Blind assessment design. Assessors conducting objective dimension scoring (Accuracy, Hallucination Resistance, Manipulation Resistance, Consistency) must not know which provider's system they are assessing until after their scores are cryptographically committed. Provider identity is revealed to objective-dimension assessors only for the purpose of preparing the assessment report after scoring is complete. This blind design is documented and verified by the Assessment Committee for each assessment cycle.
4.2 Financial independence
§ 4.2.1The Foundation maintains an operational reserve equal to at least twelve (12) months of operating expenses, invested in assets that are not dependent on the AI industry's financial health. This reserve is the material foundation of financial independence — an organization that cannot survive the loss of any single funder is not financially independent of that funder regardless of governance structures.
§ 4.2.2Assessment fees, if charged, must be structured as flat administrative fees covering the cost of conducting the assessment. They must not vary by the assessment outcome. No provision of any assessment fee arrangement may create a financial incentive for the Foundation to produce any particular score. Fee structures are published publicly and applied uniformly.
4.3 Infrastructure independence
§ 4.3.1Any entity that develops, deploys, or commercially distributes AI systems is categorically excluded from providing cloud computing, hosting, communication, or other critical infrastructure to the Foundation. This categorical exclusion is not subject to waiver and applies regardless of the commercial attractiveness of the arrangement or the donor's otherwise clean affiliation status.
§ 4.3.2The Foundation maintains documented migration plans for all critical infrastructure components, enabling transition to alternative providers within thirty (30) days. Migration plans are tested annually. No single technology provider may supply more than thirty percent (30%) of critical infrastructure. In-kind infrastructure contributions are counted toward donor concentration limits at fair market value.
4.4 Operational independence
§ 4.4.1No assessed or assessable entity may hold a seat on any Foundation governance body, including advisory bodies. No assessed or assessable entity may review, preview, or provide input on unpublished assessment scores or methodology decisions, except through the provider notification process defined in Article X and the public comment process defined in Article XIV.
§ 4.4.2Personnel post-engagement restrictions. No Foundation employee, director, or assessor who had material involvement in any specific assessment may accept employment with the assessed provider within twelve (12) months of the assessment's publication date. Material involvement is defined as contributing to the assessment score, reviewing assessment materials, or approving publication. This restriction must be incorporated into all employment agreements and assessor contracts.
Article V — Funding and Donor Influence Restrictions
Financial independence through structural funding constraints
Design note: The provisions in this article are designed specifically to address the affiliation-laundering attack vector identified in the Foundation's Assessment System Adversarial Review — the concentration of influence by a single commercial interest distributed across multiple legally distinct but functionally affiliated entities. Simple percentage-of-revenue caps are defeated by this attack. The affiliation mapping requirement in § 5.3 is the necessary counter-measure.
5.1 Permitted funding sources
§ 5.1.1The Foundation may accept funding from: philanthropic foundations not affiliated with AI developers; individual donors not currently employed by or holding equity in AI developers above the de minimis threshold; government grants from agencies that do not regulate AI developers in a manner creating regulatory conflict; academic institutions with no material commercial relationship with AI developers; and assessment or licensing fees as defined in § 4.2.2.
§ 5.1.2AI developers and AI system providers may contribute to the Foundation only through a designated Industry Contribution Program governed by § 5.5. No AI developer funding outside the Industry Contribution Program is permitted.
5.2 Concentration limits
§ 5.2.1Single-source cap. No single donor — including all affiliated entities as defined in § 5.3 — may provide more than fifteen percent (15%) of the Foundation's annual operating budget. This cap is measured on a trailing twelve-month basis and is monitored quarterly.
§ 5.2.2Industry aggregate cap. AI developers and AI system providers collectively — including all affiliated entities as defined in § 5.3, and including all contributions through the Industry Contribution Program — may not provide more than twenty-five percent (25%) of the Foundation's annual operating budget. When the aggregate approaches this cap, new industry contributions are suspended until the trailing calculation falls below twenty percent (20%).
§ 5.2.3Assessment proximity restriction. No entity that is a candidate for assessment may provide any funding to the Foundation within twelve (12) months before or twelve (12) months after their assessment's publication date. This creates a twenty-four-month funding exclusion window around each assessment. Existing commitments that fall within this window must be suspended, with the donor notified of the reason, without penalty to either party.
5.3 Affiliation mapping
§ 5.3.1Affiliation definition. For purposes of all concentration calculations in this article, entities are treated as affiliated if they share any of: common ownership or control above five percent (5%); a director or officer serving on both entities' boards; a commercial relationship exceeding $100,000 annually in either direction; common investment from a single fund representing more than ten percent (10%) of either entity's capitalization; immediate family relationships between controlling individuals; or any relationship that a reasonable observer would conclude creates shared financial interests in the Foundation's assessment outcomes.
§ 5.3.2Annual affiliation audit. An independent accountant — not the Foundation's regular auditor, and with no other relationship with the Foundation — conducts an annual affiliation mapping review specifically tasked with identifying affiliation networks among donors. The methodology of this review and its findings are published annually. Any donor who declines to provide affiliation disclosure is treated as potentially affiliated with all other non-disclosing donors for concentration calculation purposes.
§ 5.3.3Post-hoc discovery. If affiliation concentration is discovered after the fact: the Foundation publishes the finding and the gap in detection that allowed it within thirty (30) days; returns excess funding above the aggregate limit; implements enhanced disclosure requirements; and obtains independent Board member certification that no governance decisions during the concentration period were materially influenced by the affiliated donor group. The post-hoc discovery and remediation is published in the annual transparency report with full detail.
5.4 Earmarking and restricted gifts
§ 5.4.1The Foundation accepts only unrestricted gifts and gifts restricted to broad programmatic categories (e.g., "AI assessment research," "public education on AI"). No gift may be restricted to: a specific assessment of a named AI system or provider; a specific outcome, score, or finding; a specific methodology decision; or any other condition that creates a financial incentive tied to assessment conduct. Gifts offered with prohibited restrictions are declined.
§ 5.4.2All gift agreements are reviewed by the Executive Director and a Board designee before acceptance. Any gift agreement that a reasonable person could interpret as creating an improper incentive is declined regardless of the donor's stated intent.
5.5 Industry Contribution Program
§ 5.5.1The Foundation may operate an Industry Contribution Program allowing AI developers to contribute to a pooled, unrestricted fund supporting methodology development and assessment infrastructure. Contributions to this program are subject to all concentration limits, the assessment proximity restriction, and the affiliation mapping requirement. Contributors to the program have no special access to Foundation governance, no input into methodology decisions, no advance notice of assessments, and no ability to influence score publication.
§ 5.5.2Participation in the Industry Contribution Program is not listed as an endorsement of the Foundation by the contributor, and non-participation does not affect assessment eligibility, timing, or treatment. The existence of the program and all contributing entities are published publicly.
5.6 Financial transparency
§ 5.6.1The Foundation publishes on its website, updated quarterly: a list of all donors who have contributed more than five hundred dollars ($500) in the trailing twelve months, identified by name; the amount contributed by each listed donor; each donor's sector classification; a confirmation of affiliation status; and a calculation showing current concentration percentages. This disclosure is permanent — it is not removed when donor relationships end.
Article VI — Anti-Capture Mechanisms
Structural defenses against institutional capture
This article operationalizes the five structural redesigns identified in the Foundation's Assessment System Adversarial Review: cryptographic transparency, distributed authority, adversarial representation, asymmetric cost design, and behavioral monitoring. These are structural mechanisms, not policies. Policies can be waived; structures resist waiver.
6.1 Cryptographic transparency
AC-001 — Cryptographic Commitment Protocol
Every assessment creates a cryptographic commitment at the time of assessment: the test queries, the AI system responses, the reviewer scores, adjudicator assignments, and the timestamp of each action are hashed and appended to a public, append-only transparency log before the composite score is computed. This commitment proves that scores were not altered after computation, that test items were not selectively removed after responses were reviewed, and that adjudicator assignments were determined before any party was notified. Independent parties can verify these commitments against the transparency log. The log does not require publishing the test items — it commits to cryptographic hashes of those items, enabling verification without enabling gaming.
§ 6.1.1No assessment score may be published unless the cryptographic commitment for that assessment is already present in the public transparency log. The Assessment Committee verifies this condition before approving publication. The cryptographic hash of the assessment package must be published alongside every score.
6.2 Distributed authority
AC-002 — Multi-Party Authorization Requirements
The following actions require authorization from at least two independent governance bodies before they can be executed: publication of any assessment score; suspension or withdrawal of any published score; modification of a published score; adoption of any change to the IAF scoring rubrics; and exclusion of any provider from assessment eligibility. No single officer, director, or committee can execute any of these actions unilaterally.
§ 6.2.1Score publication requires: Assessment Committee approval (majority vote) and Executive Director sign-off. Score withdrawal requires: Assessment Committee approval plus Board notification with 14-day review period (Board may override withdrawal with supermajority). Score modification requires: full re-assessment under the correction process in § 9.6. Methodology changes require: public comment period, SAP review, and Board supermajority approval under Article XIV.
6.3 Adversarial representation
AC-003 — Permanent Adversarial Function
The Permanent Adversarial Function (PAF) described in § 2.5 is a standing structural mechanism, not an occasional external review. The PAF is funded from a dedicated budget line that cannot be reduced below two percent (2%) of operating expenses without a Board supermajority, ensuring it cannot be defunded as an austerity measure. The PAF's quarterly reports are published without prior review or approval by Foundation management — they go directly from the PAF to the Board Chair and are simultaneously published publicly.
§ 6.3.1The PAF's scope includes at minimum: attempting to identify weaknesses in the reviewer conflict-of-interest detection system; attempting to model the affiliation-laundering attack on Foundation funding; testing whether assessment protocols can be gamed by studying published methodology; identifying coordination patterns among reviewers that might indicate organized capture; and evaluating whether the Foundation's actual published scores are consistent with the methodology it describes. The PAF documents all findings, including findings that no significant vulnerabilities exist.
6.4 Reviewer behavioral monitoring
AC-004 — Statistical Anomaly Detection
The Foundation continuously monitors for: systematic scoring patterns inconsistent with a reviewer's claimed expertise; topic-correlated scoring that suggests ideological rather than methodological assessment; coordination patterns indicating organized reviewer campaigns; and scoring velocity anomalies (reviews submitted too quickly to represent genuine evaluation). This monitoring operationalizes Corroboration Standard safeguards CS-S-001 through CS-S-008. Detected anomalies trigger behavioral audit, not automatic disqualification — the reviewer may be correct and the anomaly may reflect genuine methodological disagreement. But the anomaly investigation must be completed and documented.
6.5 Asymmetric cost design
§ 6.5.1The Foundation designs its dispute processes to impose costs on actors attempting to abuse them. This includes: refundable bonds for institutional dispute filers that are forfeited if the dispute is found to be frivolous or coordinated; maximum active dispute limits per filer (three simultaneous disputes); and minimum cooldown periods between dispute submissions against the same assessed system (sixty days). These provisions implement the asymmetric cost principle from the Adversarial Review — legitimate disputants incur these costs once; abusive dispute flooding is deterred by its own cumulative cost.
6.6 Scientific classification protection
§ 6.6.1The IAF's distinction between settled scientific questions (where false balance is a scoring failure) and genuinely contested questions (where balance is required) is determined by the Scientific Advisory Panel, not by Foundation management or the Board. SAP determinations on scientific question classification are published with full reasoning and supporting citations. This protection prevents political capture of the Foundation's Fairness and Viewpoint Balance scoring through demands to reclassify settled science as contested — a recognized attack vector documented in the Adversarial Review.
§ 6.6.2The Foundation's position on scientific consensus questions is derivative of scientific consensus, not originary. When challenged on a settled/contested determination, the Foundation publishes the SAP's determination and its basis, and where possible engages scientific institutions to confirm the determination. The Foundation does not treat political advocacy against a scientific classification as evidence that the classification is wrong.
6.7 Legal defense preparedness
§ 6.7.1The Foundation maintains a dedicated legal defense fund separate from operating reserves, sufficient to sustain at minimum two simultaneous SLAPP suits through conclusion without material impairment to operations. The existence and approximate size of this fund is published annually. This transparency is itself a deterrence mechanism — knowing the Foundation can absorb litigation costs reduces the strategic value of SLAPP actions against assessment findings.
§ 6.7.2The Foundation will not settle any litigation on terms that require: retraction of a methodologically sound score; modification of published methodology to avoid future litigation exposure; non-disclosure of settlement terms; or any restriction on the Foundation's future assessment activities. These prohibitions are adopted in advance because the pressure to accept such terms is highest at the moment of litigation, when institutional courage is most difficult and most necessary.
Article VII — Assessment Conduct and Methodology
Standards governing how assessments are conducted
7.1 Published methodology requirement
§ 7.1.1No assessment score may be published under a methodology that has not been publicly available for at least sixty (60) days prior to the assessment cycle in which it is used. This requirement prevents retroactive methodology adoption and ensures that providers can understand the basis on which their systems will be evaluated. Emergency corrections to published methodology to address active harm require Board notification and public disclosure within twenty-four (24) hours of the correction.
7.2 Assessment protocol
§ 7.2.1Each assessment cycle begins with an Assessment Protocol document approved by the Assessment Committee, specifying: the version of the IAF and supporting standards in use; the specific prompts or evaluation items for this cycle (maintained under security protocols of § 7.3); the composition of the assessor pool; the blind assessment procedures to be applied; the confidence level to be assigned based on sample size and methodology; and the timeline from assessment commencement to score publication. The Assessment Protocol is committed to the cryptographic log at the start of the cycle.
§ 7.2.2System version commitment. Before assessment begins, the provider must provide cryptographic evidence of the specific model version or API endpoint being assessed. The provider consents to quarterly post-deployment verification checks confirming the assessed version remains accessible for comparison. Providers who will not agree to these verification terms are not assessed — and their non-participation, with the stated reason, is published on the AI Assessment Index.
7.3 Benchmark security
§ 7.3.1The Foundation maintains a private reserve of test items never published, constituting at least thirty percent (30%) of the items used in any assessment cycle. At least thirty percent (30%) of published test items are rotated per assessment cycle. This prevents full benchmark optimization and is a specific mitigation for the Goodhart's Law attack vector — providers who study published methodology and optimize for it are still assessed against a portion of the benchmark that they cannot observe in advance.
§ 7.3.2The Foundation publishes a permanent, prominent warning on all published scores: "IAF scores measure performance on assessed dimensions at assessment time. They cannot guarantee equivalent performance in deployment contexts that differ from the assessment protocol. All published benchmarks are subject to optimization by assessed entities; the IAF's value lies in making gaming visible and costly rather than impossible." This warning is not subject to waiver and cannot be removed from any publication that includes or references an IAF score.
7.4 Multi-assessor requirement
§ 7.4.1Human-review dimensions (Political Balance and Viewpoint Fairness, Cultural Fairness, Wisdom and Tradeoff Reasoning, Human Dignity and User Agency, Civic Responsibility) must be assessed by a minimum of three (3) independent reviewers per dimension. Assessors' scores must not be disclosed to each other before independent submission. Inter-rater agreement statistics must be published alongside all human-review dimension scores. Dimensions where inter-rater agreement falls below a Cohen's kappa of 0.40 are published with an explicit reliability warning.
Article VIII — Reviewer Ethics Code
Obligations of all persons conducting assessments
This code applies to all assessors, reviewers, adjudicators, and Scientific Advisory Panel members who contribute to any assessment or dispute resolution activity under Foundation authority. Acceptance of a reviewer engagement constitutes acceptance of this code.
8.1 Core obligations
§ 8.1.1Accuracy over accommodation. A reviewer's obligation is to report their honest assessment of the AI system's performance against the published methodology. Pressure from any source — Foundation management, provider representatives, fellow reviewers, or the reviewer's own commercial interests — does not alter this obligation. A reviewer who believes the correct score is unfavorable to a provider must report that score.
§ 8.1.2Methodology fidelity. Reviewers apply the published IAF methodology and assessment rubrics as written. Reviewers who believe the methodology is incorrect have an obligation to report this concern through the methodology feedback process under Article XIV. They do not have authority to deviate from the published methodology in their individual assessments.
§ 8.1.3Independence maintenance. Reviewers do not communicate with each other about specific assessment assignments before submitting independent scores. Reviewers do not communicate with provider representatives about any active assessment. Reviewers do not discuss assessment assignments with anyone outside the Foundation's defined assessment team.
§ 8.1.4Mandatory reporting. A reviewer who is contacted by any person seeking to influence their assessment findings must immediately report this approach to Foundation management, regardless of whether the contact was explicit or indirect. A reviewer who discovers a conflict of interest after accepting an assignment must immediately disclose it and recuse without waiting for formal determination of the conflict's materiality.
8.2 Conflict of interest requirements
§ 8.2.1Thirty-six month lookback. All reviewers must disclose any financial relationship with any AI developer or AI system provider within the preceding thirty-six (36) months. The 36-month lookback is the same standard applied by the Foundation's Corroboration Standard and is set at this length specifically to cover multi-year consulting arrangements, equity vest periods, and advisory relationships that may create continuing alignment of interests after formal engagement ends.
§ 8.2.2Pre-review self-certification. Before each review session, reviewers complete a signed self-certification: "I confirm that I have no financial, professional, advocacy, or personal relationship with any party, entity, or interest that would affect my ability to assess this system's performance objectively and in accordance with the published methodology. I have no undisclosed conflicts. I will immediately disclose and recuse if I recognize any element of this review that creates or suggests a conflict." False self-certification is grounds for permanent disqualification and referral to any applicable professional licensing authority.
§ 8.2.3Advocacy disclosure. Reviewers who have publicly advocated for a specific position on any contested question relevant to an assigned assessment must disclose this. Advocacy disclosure does not automatically disqualify but triggers a determination by the Foundation's COI officer. Reviewers whose advocacy position would require them to assess the provider's performance against a question where they have a disclosed advocacy interest are assigned to non-conflicted reviewers for that dimension.
8.3 Prohibited conduct
§ 8.3.1Reviewers must not: accept any gift, payment, hospitality, or benefit with a value exceeding twenty-five dollars ($25) from any assessed or assessable entity; discuss employment, consulting, or other future commercial arrangements with any assessed entity during the twelve (12) months following their involvement in that entity's assessment; use confidential assessment materials for any purpose other than the assessment; disclose unpublished assessment findings to any person outside the defined assessment team; publicly advocate for specific outcomes of an assessment in which they are participating; or purchase or short-sell securities in any assessed publicly-traded entity during the sixty (60) days preceding score publication.
8.4 Post-engagement restrictions
§ 8.4.1For twelve (12) months after the publication of an assessment in which they had material involvement, reviewers may not: accept employment with the assessed entity; enter consulting arrangements with the assessed entity; serve on the assessed entity's advisory board; or publicly comment on unpublished aspects of their assessment findings. These restrictions are incorporated into all reviewer contracts and are enforceable as contractual obligations.
8.5 Enforcement
§ 8.5.1Violations of this code are investigated by the Foundation's COI officer, with findings reported to the Assessment Committee. Sanctions range from written warning to permanent disqualification, based on the severity and willfulness of the violation. Violations that constitute conflicts of interest material to a published score trigger the overturn and correction process under § 9.6. Confirmed violations are published in the Foundation's transparency log, identifying the nature of the violation without necessarily identifying the reviewer by name except where identification is required for public accountability.
Article IX — Score Publication Standards
Requirements governing all published assessment scores
9.1 Mandatory publication elements
§ 9.1.1Every published IAF score must include: (a) the composite score on a 0–100 scale, or "INVALID" designation for floor failures as defined in § 9.2; (b) the score for each of the eleven IAF dimensions, with the weighting applied; (c) the floor dimension scores with explicit indication of any dimension scoring below 40; (d) the confidence level (L1–L5) as defined in the IAF; (e) the date of assessment; (f) the IAF version under which the assessment was conducted; (g) a description of the assessed system version, including cryptographic hash commitment; (h) the assessment sample size; (i) inter-rater agreement statistics for all human-review dimensions; and (j) the cryptographic log reference for this assessment.
§ 9.1.2Scores published without any mandatory element are not valid IAF assessments. If a mandatory element cannot be provided at time of publication, the score is withheld until it can be. The permanent benchmark limitation warning required by § 7.3.2 is published adjacent to every score and may not be removed or minimized.
9.2 Floor dimension failures
§ 9.2.1The four IAF floor dimensions — Hallucination Resistance, Manipulation Resistance, Human Dignity and User Agency, and Civic Responsibility — invalidate the composite score if any floor dimension scores below forty (40). An invalidated composite is published as "INVALID — Floor Fail on [Dimension Name]" with the actual floor dimension score displayed. The composite numeric score is not displayed for an invalid assessment. This prevents an AI system from partially compensating a catastrophically low safety or dignity score with high scores on other dimensions.
§ 9.2.2Floor failures are subject to the same provider notification, response, and dispute processes as composite scores. The fact of a floor failure is not, on its own, grounds for score correction — it reflects the methodology's deliberate design. A provider seeking to challenge a floor failure must demonstrate a specific error in the scoring of the floor dimension, not merely argue that the floor threshold is too low.
9.3 Confidence levels
§ 9.3.1All published scores carry a confidence level designation (L1 through L5) reflecting the assessment's sample size, methodology depth, and replication status. Confidence levels must be published prominently alongside scores — not in footnotes — and the limitation statement for the applicable confidence level must be visible without user action. An L1 Indicative score and an L5 Definitive score are not equivalent and must never be displayed in a manner suggesting they are comparable without qualification.
9.4 Publication timing and notification
§ 9.4.1Assessed providers receive fourteen (14) days' advance notice before score publication. This notice period is provided exclusively for factual review — identifying factual errors in the assessment (e.g., the wrong system version was assessed, a specific prompt was malformed) — and not for score negotiation. The provider may submit a written factual review response during this period. The score is not modified based on provider preference or commercial impact arguments.
§ 9.4.2If a factual error identified during the notice period is confirmed, the assessment is corrected before publication and the correction is documented in the assessment record. If the identified error is disputed between the provider and the Foundation, the score is published with the disputed element noted for further review. The provider retains the right to formal dispute under Article XI after publication.
9.5 Dispute and status labeling
§ 9.5.1All published scores carry a status label reflecting their current state: Active (no open dispute); Under Review (formal dispute filed and accepted); Suspended (temporarily withdrawn pending correction or extraordinary review); Superseded (replaced by a more recent assessment); or Historical (the system version assessed is no longer in active deployment). Status labels are updated within forty-eight (48) hours of any status change. Provider response panels are published simultaneously with scores and cannot be removed while the score remains published.
9.6 Score correction
§ 9.6.1Published scores may only be modified through a formal correction process. Corrections are triggered by: confirmed factual error in score computation; discovery of an undisclosed assessor conflict of interest material to the score; confirmed assessment protocol violation; or Stage 3 dispute resolution outcome. Score modifications based on provider preference, commercial pressure, or methodological disagreement are not corrections — they are methodology revisions and are governed by Article XIV. The correction record, including the original score, the corrected score, and the basis for correction, is permanently published.
Article X — Provider Response Rights
Rights of assessed entities before and after score publication
10.1 Pre-publication rights
§ 10.1.1Assessed providers have the right to: receive fourteen (14) days' advance notice of a score before publication (§ 9.4.1); identify factual errors during the notice period (§ 9.4.1); submit a written response of up to two thousand (2,000) words for simultaneous publication with the score; and receive a written explanation of the assessment methodology applied to their system upon request.
§ 10.1.2Assessed providers do not have the right to: review assessment prompts before or after assessment; negotiate score outcomes; condition system access on favorable treatment; delay publication by requesting extensions to the notice period (except for a single, once-per-assessment fourteen-day extension on demonstrating genuine factual error requiring investigation); or submit responses that contain false factual claims about the Foundation, its methodology, or its staff.
10.2 Provider response publication
§ 10.2.1Provider responses submitted within the notice period are published simultaneously with the score, unedited, adjacent to the score. The Foundation does not editorialize on provider responses or append counter-responses. The response is labeled as a provider statement and attributed to the provider. This right exists unconditionally — a provider whose score is catastrophically low receives the same response publication right as a provider with a high score.
§ 10.2.2Provider responses may not include: the identities or personal information of assessors or reviewers; claims that are defamatory toward identified Foundation personnel; promotional material for the provider's products unrelated to the assessment findings; or representations that the provider "endorses" or "approves" the assessment. Responses that violate these limitations are returned to the provider for revision. If the revised response continues to violate these limitations, the Foundation publishes a notation that a response was submitted and revised, and the final submitted version.
10.3 Post-publication rights
§ 10.3.1After publication, assessed providers may: submit a formal dispute under Article XI within ninety (90) days; request a re-assessment at any time (re-assessments are conducted under the same protocol as initial assessments; scores from re-assessments supersede previous scores while the previous score remains in the historical record); and submit supplementary responses of up to five hundred (500) words in response to any significant update to the Foundation's published commentary on the score.
Article XI — Dispute Resolution and Appeals
Process for challenging published assessment scores
11.1 Standing and grounds
§ 11.1.1Standing. The following parties have standing to file a formal dispute: (a) the assessed provider, for any aspect of their published score; (b) any member of the public who can demonstrate a documented interest in the accuracy of the score (researchers, journalists, consumer advocates, and users of the assessed system have documented interest as a matter of law); and (c) Foundation staff, who have an internal obligation to escalate suspected errors.
§ 11.1.2Grounds for dispute. A dispute must be grounded in one or more of: (a) factual error — the assessors misapplied the rubric criteria to the specific system outputs, not merely reached a conclusion the filer disagrees with; (b) procedural violation — the assessment did not follow the published methodology in a way material to the score; (c) undisclosed conflict of interest — an assessor had a conflict that was not disclosed and that was material to their scoring; (d) material new evidence — evidence that was not available at the time of assessment and that, if known, would materially affect the score.
§ 11.1.3Not grounds for dispute. The following are not grounds for a formal dispute: disagreement with the IAF methodology when that methodology was publicly disclosed before assessment; commercial impact of the published score; the fact that competitor systems received higher scores; disagreement with the settled/contested scientific question classifications made by the Scientific Advisory Panel; and the general claim that the assessment was "unfair" without specification of a factual error or procedural violation.
11.2 Filing requirements
§ 11.2.1A formal dispute must: be submitted in writing within ninety (90) days of score publication; identify the specific dimension(s) or elements challenged; specify the ground(s) on which the dispute is based from the list in § 11.1.2; provide documentation supporting the disputed claim; and, for institutional filers, include a refundable dispute bond of two thousand five hundred dollars ($2,500) per dispute, refunded in full if the dispute is upheld or partially upheld, and forfeited if the dispute is found to be frivolous or filed in bad faith.
§ 11.2.2Dispute limits. No filer may have more than three (3) active disputes outstanding at any time. No filer may file more than one dispute against the same assessed system within any sixty (60) day period. These limits apply per filer — entities attempting to circumvent these limits through the use of affiliated entities are subject to the affiliation definition in § 5.3.1.
11.3 Three-stage process
§ 11.3.1Stage 1 — Administrative Review (14 days). Foundation staff review the dispute for standing, timeliness, and the presence of at least one valid ground. Disputes that are procedurally deficient are returned to the filer with notice of deficiency and a fourteen (14) day opportunity to cure. Disputes that lack any valid ground are dismissed with a written explanation. Disputes that present a prima facie valid ground advance to Stage 2. All Stage 1 outcomes are published in the Foundation's transparency log within five (5) days of the determination.
§ 11.3.2Stage 2 — Assessment Panel Review (30 days). A three-person panel is constituted: one senior assessor with expertise in the relevant dimension(s), one subject matter expert independent of the Foundation, and one member of the Independent Review Panel. The panel reviews the specific claim, the assessment record, and supporting documentation. The panel may conduct supplementary assessment of the disputed dimension(s) where necessary. The panel produces a written determination on each disputed claim, with the factual basis for each finding. If the panel finds for the filer on any claim, it specifies the correction required. Panel determinations are published in full within five (5) days of issuance.
§ 11.3.3Stage 3 — External Arbitration (60 days). Either party may escalate to external arbitration after Stage 2 if the Stage 2 determination is disputed on procedural grounds (not merely substantive disagreement). The arbitrator is an independent expert agreed to by both parties from a Foundation-maintained panel of qualified arbitrators; if the parties cannot agree, the Foundation selects from the panel by lottery. Arbitration is conducted under the rules of a recognized arbitration institution. The arbitrator's determination is binding and may not be further appealed internally. The full arbitration record, including the arbitrator's reasoning, is published within thirty (30) days of issuance.
11.4 Score status during dispute
§ 11.4.1A score remains published during the dispute process, labeled "Under Review — Formal Dispute Filed." Scores are not suspended during dispute unless the Assessment Committee determines, by majority vote, that the dispute raises a credible concern about a material procedural violation that, if confirmed, would require a complete re-assessment. The bar for suspension during dispute is high precisely to prevent the dispute process from being used as a tool to suppress valid scores while disputes are pending.
Article XII — Transparency Requirements
What the Foundation must publish and when
Transparency under this charter is not a communications strategy. It is the primary mechanism by which external stakeholders can verify that the Foundation is doing what it says it is doing. The provisions in this article are minimum requirements, not ceilings.
12.1 Methodology transparency
§ 12.1.1The Foundation publishes and maintains publicly accessible: the current version of the IAF with all dimension definitions, weights, scoring rubrics, confidence level definitions, and floor dimension thresholds; the current version of the Corroboration Standard; the Assessment Charter (this document); all previous versions of each document with effective dates and change logs; the Assessment Protocol for each completed assessment cycle (with test item content redacted for benchmark security, but all structural elements published); and the cryptographic transparency log in a format accessible to independent verification.
12.2 Financial transparency
§ 12.2.1The Foundation publishes quarterly: a list of all donors above the $500 annual threshold with amounts and sector classification; current revenue concentration calculations; confirmation of compliance with all concentration limits; the results of affiliation monitoring; and a statement from the Executive Director attesting to the accuracy of the disclosures. Annually, the Foundation publishes audited financial statements prepared by an independent CPA, filed IRS Form 990 (unredacted), and the affiliation audit report required by § 5.3.2.
12.3 Conflict disclosure
§ 12.3.1Every published score includes: a statement confirming that all assessors for this assessment disclosed no material conflicts; or, if a conflict was disclosed and determined non-material, a description of the conflict and the basis for the non-materiality determination; or, if a conflict was discovered post-publication, the correction triggered under § 9.6. The identities of assessors are not published, but the domain expertise and independence status of the assessor pool is published for each assessment.
12.4 Decision transparency
§ 12.4.1The Foundation publishes within thirty (30) days: the outcome of every Stage 1 dispute triage (with the basis for dismissal or advancement); every Stage 2 panel determination in full; every Stage 3 arbitration determination in full; every score correction with the original score, corrected score, and basis; every Board decision on major matters as defined in § 3.4.1; and every PAF quarterly report. These publications are permanent — they are not removed when the underlying matter is resolved.
Article XIII — Public Accountability and Reporting
Ongoing accountability to the public and external stakeholders
13.1 Annual Transparency Report
§ 13.1.1The Foundation publishes an Annual Transparency Report within ninety (90) days of each fiscal year end. The report includes: a complete accounting of all assessments conducted during the year; all scores published with current status; a summary of all disputes filed, their grounds, and outcomes; all score corrections with basis; all conflicts detected and resolved; funding composition and concentration compliance; PAF findings summary and management response; a candid assessment by the Executive Director of where the Foundation's performance fell short of its stated standards; and a statement by the Board Chair on governance compliance.
§ 13.1.2The Executive Director's candid assessment section is not subject to Board review before publication. The requirement that it be candid — including acknowledging shortcomings — is itself a governance mechanism. An Annual Transparency Report that contains no self-criticism across any dimension signals either perfect performance (implausible) or institutional self-protection (a governance failure).
13.2 External audit
§ 13.2.1Every three (3) years, the Foundation commissions an independent external governance audit by an organization with no financial relationship with the Foundation, assessed providers, or major donors. The audit evaluates compliance with this charter's requirements across all articles. The full audit report is published without redaction except for legally privileged communications. The Board's response to audit findings, including a timeline for addressing any identified deficiencies, is published within sixty (60) days of the audit report.
13.3 Stakeholder engagement
§ 13.3.1The Foundation holds at minimum one (1) public stakeholder meeting per year at which any member of the public may submit questions in writing. Written questions are answered in writing and published. The Foundation maintains a public comment channel for ongoing feedback on methodology, governance, and published scores. Public comments and Foundation responses are published quarterly. The Foundation does not treat volume of public comment as evidence of the merits of any position — but it treats sustained, substantiated public concern as a trigger for formal review.
13.4 Error correction culture
§ 13.4.1The Foundation treats public correction of errors as a governance obligation, not a reputational risk. When an error is identified — by internal review, external dispute, or public reporting — the correction is published prominently, not buried. The original erroneous content is not deleted; it is labeled as corrected with the date of correction and the nature of the error. The Foundation does not use language that minimizes errors ("clarification," "updated information") when the accurate description is "correction." An institution that cannot acknowledge its errors cannot be trusted to accurately describe other institutions' errors.
Article XIV — Methodology Revision
Process for updating the IAF and related standards
§ 14.1Version control. All methodology documents — the IAF, the Corroboration Standard, the Pilot Benchmark, and this charter — are version-controlled. Version numbers follow a major.minor structure. Major versions (e.g., v1.0 to v2.0) reflect substantive changes to dimension weights, floor dimension thresholds, scoring rubrics, or governance requirements. Minor versions (e.g., v1.0 to v1.1) reflect clarifications, additions to non-binding guidance, or corrections of non-material errors. All versions are permanently archived and accessible.
§ 14.2Major revision process. A major revision to any methodology document follows this sequence: (1) Draft revision published for a sixty (60) day public comment period; (2) Written comments received and published in a public comment register; (3) Scientific Advisory Panel review of any changes affecting scientific question classifications; (4) Assessment Committee review of operational implications; (5) Board review and supermajority approval; (6) Publication sixty (60) days before the first assessment cycle in which the revision takes effect. A major revision may not take effect retroactively — scores published under prior versions are not recalculated under new versions.
§ 14.3Emergency methodology corrections. A clear factual error in a published methodology document — not a disagreement with a methodology choice, but an actual error — may be corrected through an expedited process: Board notification, twenty-four (24) hour review period, publication with a clear "Correction" designation. Emergency corrections may not change dimension weights, floor thresholds, or governance requirements. They may correct mathematical errors, typographical errors that affect scoring calculations, and cross-references that point to incorrect provisions.
§ 14.4Benchmark rotation. Test item rotation under § 7.3.1 occurs on an administrative cycle and does not require the major revision process. The existence and approximate scope of rotation is disclosed in each Assessment Protocol; the specific items added or removed are not disclosed, as that information would enable gaming of the items remaining in use.
§ 14.5Methodology feedback from reviewers. Reviewers who identify potential errors or weaknesses in the published methodology during the course of an assessment have an obligation to report these concerns through a designated feedback channel. Feedback is aggregated, reviewed by the Assessment Committee quarterly, and addressed either through the revision process or through a published explanation of why no change is warranted. Reviewer feedback is a formal input to methodology improvement — it is not merely accommodated.
Article XV — Amendment and Interpretation
How this charter changes and how it is read
§ 15.1Amendment authority. This charter may be amended only by a two-thirds (2/3) supermajority of the full Board of Directors (not merely of a quorum), following a sixty (60) day public notice period during which the proposed amendment is published and public comment accepted. The charter may not be amended to: eliminate the independence supermajority requirement in § 3.1.2; reduce the funding concentration limits below the thresholds in § 5.2; eliminate the cryptographic transparency requirement in § 6.1; eliminate the Permanent Adversarial Function; or reduce provider response rights below those in Article X. These provisions are entrenched and require a separate, additional referendum process defined in Schedule D before they can be modified.
§ 15.2Interpretation principle. When any provision of this charter is ambiguous, the interpretation that best advances the Founding Principles in the Preamble governs. In particular: interpretations that favor institutional convenience over independence safeguards are disfavored; interpretations that favor transparency over privacy (except for individual reviewer identities) are preferred; interpretations that favor the long-term credibility of the assessment function over short-term operational ease are preferred.
§ 15.3Conflict with other documents. This charter supersedes all prior governance documents of the Foundation on matters within its scope. Where this charter is silent on a matter governed by the IAF, Corroboration Standard, or other published Foundation standards, those documents govern their respective domains. In the event of a direct conflict between this charter and any other Foundation document, this charter governs.
§ 15.4Effective date. This charter takes effect upon adoption by the Board of Directors and supersedes all prior governance arrangements on matters within its scope. Assessment cycles commenced before this charter's effective date are governed by the governance arrangements in effect at commencement. All future cycles are governed by this charter.
Schedules and Cross-References
Supporting materials and document relationships
Schedules referenced in this charter
| Schedule | Subject | Status |
|---|
| Schedule A | Independent Review Panel — Composition, appointment, rotation, and compensation | To be adopted by Board resolution within 90 days of charter adoption |
| Schedule B | Scientific Advisory Panel — Composition, appointment, term, and classification procedure | To be adopted by Board resolution within 90 days of charter adoption |
| Schedule C | Permanent Adversarial Function — Charter, scope, staffing, and reporting requirements | To be adopted by Board resolution within 90 days of charter adoption |
| Schedule D | Entrenchment referendum process — For modification of entrenched provisions in § 15.1 | To be adopted by Board resolution within 90 days of charter adoption |
| Schedule E | Arbitrator panel — Qualified arbitrators for Stage 3 dispute resolution under § 11.3.3 | To be developed within 180 days of charter adoption |
Primary document cross-references
| Document | Relationship to this charter |
|---|
| Intelligence Assessment Framework v1.0 | Defines the assessment methodology; this charter governs the governance structures within which the IAF is applied. IAF governs assessment conduct; this charter governs organizational independence, publication, dispute, and accountability |
| Corroboration Standard v1.0 | Governs reviewer qualification, blind review protocol, conflict of interest procedures, and the Reviewer Integrity Score. This charter's Article VIII (Reviewer Ethics Code) supplements and does not replace the Corroboration Standard's reviewer provisions |
| IAF Pilot Benchmark v1.0 | Provides the specific assessment prompts and rubrics for IAF assessment cycles; governed by methodology security provisions in § 7.3 |
| Assessment System Adversarial Review | Identifies the threat landscape that this charter's anti-capture mechanisms (Article VI) are designed to address; published openly as a transparency measure |
| AI Assessment Index | The public-facing publication of assessment scores; governed by score publication standards in Article IX and provider response rights in Article X |
Glossary of defined terms
| Term | Definition in this charter |
|---|
| Assessment Independence | Defined in § 4.2 — no financial relationship with any assessed or assessable entity in the preceding 36 months |
| Affiliated entities | Defined in § 5.3.1 — entities sharing common ownership, board membership, material commercial relationships, common investment, or family relationships between controlling individuals |
| Floor dimensions | IAF dimensions whose score below 40 invalidates the composite score: Hallucination Resistance, Manipulation Resistance, Human Dignity and User Agency, Civic Responsibility |
| Material conflict | Any conflict that a reasonable observer would conclude could have influenced the assessor's scoring, regardless of the assessor's subjective belief about its influence |
| Settled scientific question | A question for which there is scientific consensus as determined by the Scientific Advisory Panel under § 6.6 — where presenting false balance is itself a scoring failure |
| Assessment cycle | An annual assessment period defined in the Assessment Protocol; the unit of measurement for assessor rotation requirements |
Adoption Record
This charter was adopted by the Board of Directors of EM Foundation by the requisite supermajority vote on the date indicated below. It constitutes the binding governance instrument for all assessment activities of the Foundation from its effective date.
Chair, Board of Directors
Board Vote (For / Against / Abstain)